bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
1 Emission rates of species-specific volatiles change across communities of Clarkia species:
2 Evidence for character displacement in floral scent.
3 Katherine E. Eisen1,2, Monica A. Geber1, Robert A. Raguso3
4 1 Department of Ecology and Evolutionary Biology, Cornell University, Ithaca NY,
5 14853, USA.
6 2 Email for correspondence: [email protected]
7 3 Department of Neurobiology and Behavior, Cornell University, Ithaca NY, 14853,
8 USA.
9 Keywords: allopatry, sympatry, multimodal floral traits, context-dependent selection
10 Word count: 5,864 words
11 This manuscript contains four figures and one table in the main text, two methodological
12 appendices, and six supplemental tables.
13 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
14 Abstract
15 A current frontier of character displacement research is to determine if displacement occurs via
16 multiple phenotypic pathways and varies across communities with different species
17 compositions. Here, we conducted the first test for context-dependent character displacement in
18 multimodal floral signals by analyzing variation in floral scent in a system that exhibits character
19 displacement in flower size, and that has multiple types of sympatric communities. In a
20 greenhouse common garden experiment, we measured quantitative variation in volatile emission
21 rates of the progeny of two species of Clarkia from replicated communities that contain one,
22 two, or four Clarkia species. The first two axes of a constrained correspondence analysis, which
23 explained 24 percent of the total variation in floral scent, separated the species and community
24 types, respectively. Of the 23 compounds that were significantly correlated with these axes, nine
25 showed patterns consistent with character displacement. Two compounds produced primarily by
26 C. unguiculata and two compounds produced primarily by C. cylindrica were emitted in higher
27 amounts in sympatry. Character displacement in some volatiles varied across sympatric
28 communities and occurred in parallel with displacement in flower size, demonstrating that this
29 evolutionary process can be context-dependent and may occur through multiple pathways. bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
30
31 Introduction
32 Interspecific interactions have long been hypothesized to have significant effects on
33 patterns of biodiversity (Darwin 1859; Lack 1945; Schluter 2000; Grant and Grant 2008). One
34 such evolutionary process is character displacement, which leads to a pattern of differences in
35 species’ trait values in sympatric communities relative to allopatric communities (Brown and
36 Wilson 1956; Germain et al. 2017). While character displacement has been studied and debated
37 for over sixty years (Stuart and Losos 2013), there are two key gaps in our understanding of this
38 process. First, outside of a small number of classic systems (e.g., anoles, sticklebacks, Darwin’s
39 finches), few studies have examined the potential for character displacement in more than one
40 type of trait (reviewed in Stuart and Losos 2013). Examining variation in multiple types of traits
41 increases our ability to detect non-repeatable character displacement, which may occur through
42 different phenotypic pathways across communities (Losos 2011; Germain et al. 2017), and
43 determine when species interactions lead to shifts in correlated or independently-evolving traits.
44 Second, while most studies of character displacement have focused on pairwise interactions (but
45 see Lemmon and Lemmon 2010; Miller et al. 2014a; Grant 2017; Roth-Monzon et al. 2020),
46 many species exist in complex ecological communities, where interactions with multiple species
47 could include indirect and higher-order interactions (Mayfield and Stouffer 2017; TerHorst et al.
48 2018; Roth-Monzon et al. 2020). Testing for character displacement across sympatric
49 communities that vary in species composition or richness (Eisen and Geber 2018; Roth-Monzon
50 et al. 2020) can advance our understanding of the evolutionary consequences of direct and
51 indirect interactions (Walsh 2013; TerHorst et al. 2015). bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
52 Among co-occurring flowering plants, pollinators often represent a shared resource that is
53 critical for reproduction (Waser et al. 1996; Ollerton et al. 2011), and there is a growing body of
54 evidence for character displacement in plants mediated by interactions between co-occurring
55 species that share pollinators (reviewed in Beans 2014; Eisen and Geber 2018). Presently, there
56 are two critical gaps in our understanding of this process. First, studies to date have examined
57 character displacement in floral morphology, color, and phenology (reviewed in Beans 2014;
58 Eisen and Geber 2018), which reflects a general bias towards visual traits in pollination (Raguso
59 2008a). Nonetheless, olfactory and reward traits are critical to successful pollination in many
60 systems (Schiestl 2010, 2015; Raguso 2014) but have yet to be integrated in to the study of
61 character displacement. Second, character displacement in floral traits is likely to occur via
62 multiple phenotypic pathways or changes in trait combinations across communities (Losos 2011;
63 Germain et al. 2017). Nonetheless, most studies to date have addressed character displacement in
64 one type of floral trait (e.g., color or morphology), not the multi-modal bouquet flowers typically
65 present (Leonard et al. 2011).
66 The emission of floral scent—volatile organic compounds including monoterpenes,
67 sesquiterpenes, and aromatic compounds (Knudsen et al. 2006)—is a complex trait, in that
68 individual plants can exhibit qualitative variation in the blend of volatiles and quantitative
69 variation in their emission rates (Raguso 2008b). Because scent can be produced not only from
70 petals but also from reproductive floral structures, scent may be correlated with or unrelated to
71 variation in flower size (Effmert et al. 2006; Valdivia and Niemeyer 2006; Burdon et al. 2015;
72 Martin et al. 2017), which could lead to multi-modal character displacement in some systems. In
73 addition, species may vary in common volatiles that are produced by a small number of
74 biosynthetic pathways (Dudareva and Pichersky 2006), and in species-specific volatiles that bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
75 provide ‘private channels’ for communication with specialist pollinators (Raguso 2008b; Soler et
76 al. 2010). Insights from three areas of floral scent research suggest that floral scent could
77 undergo character displacement. First, floral scent exhibits substantial intraspecific variation
78 across populations in multiple systems, including cacti (Schlumpberger and Raguso 2008),
79 cycads (Suinyuy et al. 2012), saxifrages (Friberg et al. 2019), and orchids (Gross et al. 2016;
80 Chapurlat et al. 2018). These patterns suggest that floral scent may be relatively evolutionarily
81 labile. As a result, scent could evolve in response to geographic variation in selection (Gross et
82 al. 2016), which could lead to variation in character displacement across different communities
83 (Germain et al. 2017; Eisen and Geber 2018). Second, floral scent can be a target of pollinator-
84 mediated selection (Parachnowitsch et al. 2012; Chapurlat et al. 2019), which indicates that floral
85 scent could evolve in response to interactions between co-occurring plant species that share
86 pollinators. Third, differences in floral scent can mediate reproductive isolation between co-
87 occurring species (Waelti et al. 2008; Bischoff et al. 2014; Peakall and Whitehead 2014) and
88 explain variation in the structure of plant-pollinator networks (Junker et al. 2010; Larue et al.
89 2016; Kantsa et al. 2018, 2019). As such, floral scent may determine how pollinators are
90 partitioned among co-occurring plant species.
91 In this study, we test for variation in multimodal character displacement across sympatric
92 communities that contain different numbers of co-occurring species. Specifically, we assess the
93 potential for character displacement in the floral scent of two co-occurring species of California
94 native annuals in the genus Clarkia (Onagraceae). These species, C. unguiculata and C.
95 cylindrica, co-occur more frequently than expected by chance in the southern foothills of the
96 Sierra Nevada (Kern County, CA, USA). Where they co-occur, these species have converged in
97 flowering time and diverged in flower size (Eisen and Geber 2018). This pattern of divergence in bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
98 flower size provides an opportunity to test if character displacement occurs on multimodal floral
99 signals. We conducted a greenhouse common garden experiment to measure quantitative
100 variation in volatile emission rates of the progeny of plants from natural communities that
101 contain one, two, or four Clarkia species. By eliminating variable environmental effects on trait
102 values, the common garden enabled us to compare phenotypes across different, replicated
103 community types and test for significant interactions between species and community types on
104 floral volatile emission rates. These data were used to address three questions regarding the
105 potential for and nature of character displacement in floral scent:
106 (1): Is variation in volatile emissions across species and community types consistent with
107 character displacement?
108 (2): Do patterns of character displacement vary across types of sympatric communities?
109 (e.g., two-species vs. four-species communities)
110 (3): Do multi-modal signals (e.g., floral scent and flower size) jointly undergo character
111 displacement?
112 Methods
113 Study system. Species in the genus Clarkia (Onagraceae) often co-occur and share pollinators,
114 which are primarily solitary bee pollinators that specialize on the genus (Lewis 1953; MacSwain
115 et al. 1973; Singh 2014). Across the genus, species exhibit intra- and interspecific variation in
116 multiple types of floral traits, including flowering time (Lewis 1961; Jonas and Geber 1999;
117 Moeller 2004; Gould et al. 2014; Singh 2014), floral orientation (Lewis 1961), petal coloration
118 (Lewis and Lewis 1955), flower size (Eisen and Geber 2018), and floral scent (Miller et al.
119 2014b). In the Southern Sierra Nevada (Kern River Canyon, Kern County, CA), C. unguiculata
120 Lindley and C. cylindrica ssp. clavicarpa W. Davis co-occur more frequently than expected by bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
121 chance (Eisen and Geber 2018). These species are primarily outcrossing because flowers are
122 protandrous and herkogamous, and while they share pollinators (Singh 2014), they are not
123 known to hybridize in the field (MacSwain et al. 1973). The petal area of C. cylindrica exhibits
124 divergent character displacement (an increase in petal area) relative to C. unguiculata in
125 communities that contain two and four species of Clarkia (Eisen and Geber 2018).
126 Common garden source community selection. Our common garden contained three replicates of
127 each of four unique types of source communities: C. cylindrica alone, C. unguiculata alone, C.
128 cylindrica and C. unguiculata together, and these two species with the two other outcrossing
129 Clarkia species (C. speciosa and C. xantiana) that occur in the Kern River Canyon (for
130 community locations, see Table S1). In other words, individuals of each species (C. cylindrica
131 and C. unguiculata) were grown from seeds sourced from three single-species communities,
132 three two-species communities, and three four-species communities. Community types thus vary
133 in how many species are present in the community. Seeds from both species were collected at all
134 communities in 2017. Three or more fruits per plant were collected from 50 haphazardly chosen
135 plants of each species at each community. The seeds from one fruit from each of 20 plants per
136 community and species were combined to ensure that plants in the common garden represented a
137 sufficient range of any possible plant-level variation at each community.
138 Plant germination and growth. Because of the large number of community x species
139 combinations present in the common garden, seeds were started in five batches in September-
140 November 2017. All community x species combinations were included in each batch of plants.
141 To break dormancy, seeds were placed on moist filter paper in a petri dish, wrapped in parafilm,
142 and stratified at 5°C for five to seven days and then held at 23°C for five to seven days before
143 planting. Germinants were transplanted into 656 ml3 Cone-tainers (Stuewe & Sons, Tangent, bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
144 Oregon, USA) filled with Lambert soil mix. The pots’ positions on the greenhouse bench were
145 randomized. Plants were exposed to supplemental light (16 h days) and maintained at 23-25°C
146 during the day and 19-21°C at night. Plants were watered twice a week on average and received
147 on average 30-40 mL of water per week in weeks 1-3 post transplanting, 70-80 mL per week in
148 weeks 4-6, and 100 mL per week in weeks 7-10. Each pot initially contained two germinants;
149 pots were thinned after four weeks to contain one plant. At this time, six prills of Osmocote®
150 Smart-Release® Plant Food Flower & Vegetable 14-14-14 fertilizer (Scotts Miracle-Gro
151 Company, Marysville, OH) were applied to the soil surface in each pot.
152 Qualitative scent analysis. To inform our quantitative sampling protocols, we conducted two
153 types of qualitative analyses using Solid Phase Micro Extraction (SPME) fibers (Supelco, Inc.,
154 (Sigma-Aldrich), Bellefonte, PA) (Appendix 1). First, to determine if the presence of additional
155 flowers changed the composition of the volatile profile (i.e., due to threshold dosage effects), we
156 compared the profiles of samples with three versus six cut flowers from the same plant. We
157 recovered significantly more monoterpenoid and sesquiterpenoid compounds in samples with six
158 flowers (Appendix 1). Given this result, we adjusted our quantitative headspace sampling
159 protocol to include a minimum of six open flowers per plant (see below). Second, to determine
160 where volatiles are produced in these flowers, we compared the volatile profiles of dissected
161 petals from six flowers versus those of the remaining tissues of the same six flowers. We found
162 that petals generally contained fewer volatiles than the non-petal floral tissues (Appendix 1),
163 which may influence the relationship between flower size and floral scent (see Discussion).
164 Quantitative scent analysis. Floral volatile samples were collected using the dynamic headspace
165 adsorption technique between November 13, 2017, and February 5, 2018. All collections were
166 made under natural lighting conditions in a well-aerated glassed-in corridor adjacent to the bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
167 greenhouse where the plants were grown. We used an AIRCARE hygrometer (Essick Air
168 Products, Little Rock, AR, USA) to record the minimum and maximum temperature and percent
169 humidity during sampling; the average minimums were 17 °C and 23 percent humidity, while the
170 average maximums were 25 °C and 41 percent humidity. Floral samples were obtained from
171 fifteen plants per community per species (Ntotal = 270), and one vegetative control sample was
172 collected per community per species (Ntotal = 18).
173 We used 16 oz PET water bottles to enclose stems for headspace sampling. Water bottles
174 were washed with odorless soap, dried, and baked in a clean drying oven for 15-20 mins at 80 °C
175 each morning before sampling began. Samples were collected using PAS-500 Micro Air Sampler
176 pumps (Spectrex, Inc., Redwood City, CA) connected to traps that contained 0.0100 g of Tenax
177 80/100 adsorbent (Alltech Associates, Inc. (W.H. Grace), Deerfield, IL, USA). The flow rate of
178 the pump was set to 200 mL/min. Scent was collected for 6 hours, from 900 to 1500, as this
179 corresponds to the period of greatest pollinator activity in natural communities. For floral
180 samples, 6-13 flowers were enclosed per plant (average: 7.4 flowers), and the number of flowers
181 enclosed was recorded for each plant sample. Vegetative controls were obtained from plants that
182 had not begun to flower but had formed buds. During each sampling day, one ambient control
183 sample was collected in an empty PET bottle.
184 Immediately following the end of the headspace collection period, the traps were
185 removed from the pumps and eluted with 300 µL of GC-MS quality hexane (Burdick & Jackson
186 GC2; Honeywell International, Inc. USA). Samples were then concentrated to 50 µL with a flow
187 of gaseous N2, and spiked with 23 ng of toluene (5 mL of a 0.03% solution in hexane) as an
188 internal standard in preparation for analysis with gas chromatography - mass spectrometry (GC-
189 MS; see below). Samples were stored at -20C and labeled with a community-neutral identifier bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
190 code based on the date of sampling (e.g., December 15-1, December 15-2, etc.) to facilitate
191 analysis that would be blind to the species and community type (see Becklin et al. 2011).
192 Scent analysis via GC-MS. Both solvent-eluted and solvent-free (SPME) volatile samples were
193 analyzed using a GC17A gas chromatograph coupled with a QP5000 quadrupole mass
194 spectrometer (Shimadzu Scientific Instruments, Inc., Kyoto, Japan). One µL aliquots of the
195 solvent eluted samples were injected (splitless mode) at 240C onto a polar GC column (EC Wax,
196 30m long, 0.25 mm internal diameter, 0.25µ film thickness; BGB Analytik). The GC oven
197 program (40C to 240C, increasing at 20C per minute, with a 2-minute hold at the maximum
198 temperature) was optimized to minimize run length (for over 300 samples) while allowing for
199 peak resolution to baseline. Electrical ionization mass spectra were generated under 70eV
200 conditions (scanning range 40-350 m/z), and resulting mass spectra were compared with those of
201 MS libraries (Wiley, NIST, Adams) using Shimadzu GCMSolutions software. Kovats retention
202 indices (KI) were prepared for each compound by running a blend of n-alkanes (C7-C30) under
203 the same chromatographic conditions and optimized method. Volatile compounds were
204 identified via 1) direct comparison of retention time and mass spectra with those of authentic
205 standards, 2) comparison with the KI of the best MS library fit for the unknown with published
206 KI values from the plant volatile literature (NIST WebBook; https://webbook.nist.gov/), or 3) in
207 the absence of standards or published KI values, the mass spectral data (ion fragment table) were
208 listed for unknown compounds in reverse order of abundance, starting with the base peak (set to
209 100%).
210 Extraction and processing of quantitative data.
211 Peak areas were integrated manually using Shimadzu GCMSolutions software. After
212 excluding compounds that were present in one or two samples out of 270, our quantitative bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
213 dataset contained 54 compounds (Table S2). To exclude experimental artifacts from individual
214 plants’ profiles, we compared the profile of each sample to the profile of the ambient control that
215 was collected on the same day. If a compound appeared in both a floral sample and the relevant
216 ambient control, we only retained this peak in the floral sample if the floral sample peak area was
217 at minimum five times larger than the peak area of the ambient control. This value was selected
218 to be highly conservative in terms of the compounds that we retained in samples where those
219 compounds were also present in the control. Similarly, to exclude compounds emitted by
220 vegetation, we compared each floral sample to the vegetative control collected from the sample
221 population and applied the same 5x threshold to any overlapping compounds. As such, some
222 compounds were retained in the dataset but excluded from particular samples where their
223 emission rates were similar to the relevant ambient or vegetative controls.
224 Emission rates were normalized by dividing total ion current (TIC) peak areas by that of
225 the internal standard (Svensson et al. 2005), then were calculated algebraically using response
226 factors generated using external standard dose-response curves generated from log- and semi-log
227 dilutions of the primary floral volatiles identified in these analyses ((E)-β-ocimene, α-pinene, β-
228 caryophyllene, benzyl alcohol, and methyl salicylate).
229 To relate emission rates to floral masses, we measured the fresh and dry masses of twenty
230 flowers (ten male-phase and ten female-phase flowers) per community and species (Table S1).
231 Flowers were selected haphazardly from between four and eight plants per community and
232 species. Each plant contributed a maximum of five flowers to the 20 total flowers per community
233 and species. Fresh masses were recorded immediately after removing the flower from the plant.
234 Flowers were dried for 24 hrs at 50 °C before dry masses were recorded. We present analyses of
235 floral scent emission rates that were standardized by the number of open flowers that contributed bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
236 to a sample multiplied by the average fresh mass of a flower from that community and species,
237 which gives the µg scent per g fresh floral mass per hr. Analyzing the data using emission rates
238 that were standardized by the number of open flowers that contributed to a sample (µg scent per
239 flower per hr) yielded highly similar results (results not shown).
240 Additional common garden to test for wounding artifacts. We observed differences across
241 species and community types in compounds that are generally considered “green leafy volatiles”
242 (abbreviated as GLVs) and are associated with plant wounding (Visser and Ave 1978; Scala et
243 al. 2013) (see Results). To determine if these patterns resulted from artifacts of the experimental
244 sampling process, we conducted an additional common garden experiment to test for differences
245 in the emission rates of GLVs between wounded and non-wounded plants (see Appendix 2). B
246 Wounding elevated the emission rates of GLVs (e.g., (Z)-3-hexen-1-ol, (Z)-3-hexenyl acetate),
247 but the emission rates of these compounds during the 2018 main experiment were more similar
248 to the 2019 non-wounded control plants than to the 2019 wounded plants (see Appendix 2). As
249 such, observed emission rates of these compounds are unlikely to be an experimental artifact and
250 we retain the GLVs as floral compounds in our analysis.
251 Multivariate statistical analyses and dimensionality reduction. We used multivariate and
252 univariate methods to test for significant interactions between species and community type,
253 which provides evidence for differences in species' emission rates of volatiles across
254 communities (character displacement). All analyses were performed in R (R Core Team 2018).
255 To partition the observed variance in the emission rate of all compounds across the species and
256 community types, we performed a Permutational Multivariate Analysis of Variance
257 (PERMANOVA) using a Bray-Curtis distance matrix using the adonis function from the vegan
258 package (Oksanen et al. 2018). We specified our replicate communities nested within community bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
259 type as strata in the function, which is equivalent to a random effect. To find the compounds that
260 distinguish the community types and species, we performed a Canonical Analysis of Principal
261 Coordinates with a Bray-Curtis distance dissimilarity index using the capscale function from the
262 vegan package (Oksanen et al. 2018).
263 Univariate statistical analyses. We analyzed variation in compound classes and in specific
264 compounds using general linear mixed-effects models, which were performed using the lme4
265 package (Bates et al. 2015). Models were assessed to ensure normally distributed residuals with
266 homogenous variance. These models all contained community type, species, and their interaction
267 as fixed effects, and community nested within community type was included as a random effect.
268 The significance of fixed effects in linear mixed models was assessed using the ANOVA
269 function in the lmerTest package ver. 2.0-29 (Kuznetsova et al. 2015) to perform type III F tests
270 using the Kenward-Roger approximation for the denominator degrees of freedom. When
271 ANOVAs returned significant F values, we used Tukey’s honest significant difference tests to
272 determine which group means were significantly different using the emmeans function with the
273 pairwise option in the emmeans package (Lenth 2019). These tests were performed with the 'type
274 = "response"' option such that intervals were back-transformed from the log and square-root
275 scales. Contrasts for models with log-transformed response variables are presented on the log-
276 odds scale, such that ratios greater than one indicate larger emission rates in C. cylindrica and
277 ratios less than one indicate larger emission rates in C. unguiculata.
278 We performed two types of univariate analyses. First, we tested for differences in total
279 scent emission and the emission of certain types of compounds across the species and
280 community types using linear mixed-effects models as described above. The compound classes
281 we analyzed were monoterpenoids, sesquiterpenoids, GLVs, and aromatics (Table S2). To bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
282 ensure that our models had normally distributed residuals with homogenous variance, total scent,
283 monoterpenoid, and aromatic emission rates were square-root transformed and GLVs and
284 sesquiterpenoid emission rates were log-transformed.
285 Second, we performed univariate analyses on compounds that were correlated with one
286 or both of the first two CAP axes. Specifically, 23 compounds were correlated with one or both
287 of the first two CAP axes (Table S3). These compounds had significant Pearson correlation
288 coefficients with one or both axes at P < 0.01 after applying a false discovery rate correction for
289 54 tests; all significant correlations were greater than 0.15 or less than -0.15. Most compounds
290 were either square-root or log-transformed to improve the normality of model residuals (Table
291 S3). To test for a significant interaction between species and community type, we ran ANOVAs
292 on these models as described above. We applied a false discovery rate correction for 23 tests on
293 the P values associated with these species by community type interactions. We then tested for
294 differences between species and community types using the emmeans function as described
295 above. All datasets and analysis scripts are available on github (https://github.com/kate-
296 eisen/clarkia-scent).
297 Results
298 Diversity of volatile organic compounds. There were 54 volatile organic compounds present in
299 four or more samples: 22 monoterpenoids, 18 sesquiterpenes and C15 derivatives, five GLVs and
300 nine aromatic or nitrogenous compounds (Table S2). Thirty-eight of the 54 compounds were
301 found in more than five samples of both species. Of the remaining 16 compounds, 11 compounds
302 were completely or nearly unique to C. unguiculata (present in five or fewer C. cylindrica
303 samples), and five compounds were completely or nearly unique to C. cylindrica (present in five bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
304 or fewer C. unguiculata samples). The average number of compounds detected in a sample
305 (mean ± 1 SE) was 13.8 ± 0.4 for C. cylindrica and 14.0 ± 0.5 for C. unguiculata.
306 Multivariate analyses. The PERMANOVA on the scent compounds revealed main effects of
307 community type (R2 = 0.03, P < 0.001), species (R2 = 0.22; P < 0.001), and an interaction
308 between the two (R2 = 0.04, P < 0.001).
309 The Canonical Analysis of Principal Coordinates indicated that a subset of all compounds
310 helped to define variation among the six species x community type combinations in our study
311 (two focal species x three community types per species). The constrained portion of the variance
312 was 24% of the total variance (our independent variables explained 24 % of the total variation in
313 the data). CAP axis 1 explained 20 % of the total variation in scent and 77 % of the constrained
314 variance. In general, C. cylindrica individuals had negative values on CAP axis 1, while C.
315 unguiculata individuals had positive values (Figure 1, Table S4). CAP axis 1 was strongly
316 positively correlated with two monoterpenoids and an aromatic compound, and strongly
317 negatively correlated with sesquiterpenoids (Table 1, Table S3). CAP axis 2 primarily separated
318 C. unguiculata individuals from the three different community types (Table S4). This axis
319 explained 4 % of the total variation in scent, and 14 % of the constrained variance. It was
320 strongly positively correlated with two monoterpenoids and an aromatic compound, and strongly
321 negatively correlated with two GLVs and a monoterpenoid (Table 1, Table S3).
322 Univariate analyses of total scent, compound classes, and single compounds. Patterns of
323 variation in three compound classes were consistent with character displacement: sesquiterpenes
324 (F2,117.84 = 8.749, P = 0.0003), GLVs (F2,78.42 = 17.330, P < 0.0001), and aromatics (F2,89.51 =
325 4.720, P = 0.0113). The interaction in sesquiterpene emissions was driven by a significantly
326 larger difference between the species in one-species communities relative to two-species bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
327 communities (Figure 2; Table S5). For the GLVs, Clarkia unguiculata produced more than C.
328 cylindrica at one-species communities and two-species communities; the interaction was driven
329 by both species producing equivalent amounts of these compounds at four-species communities
330 (Figure 2; Table S5). The interaction in aromatics emissions was driven by greater production by
331 C. cylindrica at two-species communities (Figure 2; Table S5), such that the difference between
332 the species at two-species communities was significantly larger than the differences between the
333 species at one- and four-species communities (Figure 2; Table S5). In particular, this pattern was
334 driven by the emission of large amounts of benzyl alcohol by C. cylindrica in two-species
335 communities (results not shown).
336 We used the results of the Canonical Analysis of Principal Coordinates to determine the
337 compounds that we analyzed individually. Of the 23 compounds that were correlated with one or
338 both of the first two CAP axes (see Methods), nine compounds had significant community type
339 by species interactions in univariate models (Table S6). Two of these compounds, 2-amino
340 phenyl ethanone, and methyl nicotinate, are primarily produced by C. unguiculata, and emission
341 rates were higher at both types of sympatric communities (Figure 3 A&B, Table S5). Two
342 additional compounds, (E)-cinnamic aldehyde and veratrole, are primarily or exclusively
343 produced by C. cylindrica, and emission rates were higher at both types of sympatric
344 communities (Figure 3 G & H, Table S5). The remaining five compounds with significant
345 interactions, α-pinene, β-pinene, sabinene hydrate, γ-terpinene, and (Z)-3-hexenyl acetate, are
346 primarily produced by C. unguiculata and had lower emission rates in four-species communities
347 relative to one- and two-species communities (Figure 3, Table S5).
348 Discussion bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
349 By measuring floral scent variation across communities that contain different numbers of
350 species in a system that exhibits character displacement in flower size, we conducted the first test
351 for context-dependent multi-modal character displacement in floral traits. These species exhibit
352 pronounced differences in their floral scent profiles, with more subtle but significant differences
353 across the community types. In an analysis of all of the volatile organic compounds emitted by
354 the two species, the significant interaction between species and community type was driven by
355 compounds that were primarily or exclusively emitted by only one species—two aromatic
356 compounds and four monoterpenoids emitted by C. unguiculata, and two aromatic compounds
357 emitted by C. cylindrica. These patterns were consistent across sympatric communities for C.
358 cylindrica but not for C. unguiculata. In addition, our investigation of the potential for multi-
359 modal character displacement revealed that changes in floral scent were associated with changes
360 in flower size in C. cylindrica but not in C. unguiculata. Here we discuss the potential drivers
361 and ecological implications of these patterns.
362 Character displacement driven by changes in species-specific volatiles. Because floral scent is a
363 complex trait, character displacement could occur through several pathways, including both
364 qualitative or quantitative changes in compounds that are either shared across the species or
365 unique to one species. In this study, we observed patterns consistent with character displacement
366 in compounds that were generally emitted by only one of the focal species. These types of
367 changes could be linked to increases in plant-pollinator specialization in multi-species
368 communities. Specifically, an increase in species-specific volatile emissions may increase a
369 pollinator’s ability to differentiate between two co-occurring plant species, which could increase
370 pollinator constancy and decrease heterospecific pollen transfer among species that share
371 pollinators (Waser 1986; Sargent and Ackerly 2008). Divergence in flower color has been bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
372 demonstrated to reduce inconstant foraging in multiple systems (Levin 1985; Hopkins and
373 Rausher 2012; Muchhala et al. 2014; Norton et al. 2015). The compounds that exhibited patterns
374 consistent with character displacement in Clarkia were benzenoid aromatics (both species) and
375 monoterpenoids (C. unguiculata). Among plants that are pollinated by food-seeking bees, scent
376 profiles are commonly dominated by benzenoids, terpenoids, or a mixture of the two types of
377 compounds (Dobson 2006). In both observational and experimental studies, benzenoids have
378 been associated with visitation from apid and halictid bees (Theis 2006; Andrews et al. 2007;
379 Kantsa et al. 2019), such that the increases in benzenoid emission rates in Clarkia could result in
380 greater attraction of these pollinator species. In particular, because only C. unguiculata receives
381 upwards of five percent of all pollinator visits from apid bees (Apis mellifera, Xylocopa
382 tabaniformis, Bombus sp.; Singh 2014), the increases in benzenoid emissions could reflect
383 greater pollinator specialization in sympatric communities.
384 Context dependency of character displacement. Because indirect interactions can modify
385 evolutionary trajectories (Benkman 2013; Walsh 2013; TerHorst et al. 2015), we tested for
386 variation in character displacement in two types of sympatric communities: two-species
387 communities that contain the focal species of this study, and four-species communities that
388 contain the focal species plus two additional congeners that flower later in the summer (Moeller
389 2004). We found that C. cylindrica exhibited similar patterns across both types of sympatric
390 communities, while patterns for C. unguiculata across the community types varied by compound
391 class (monoterpenoids and aromatics). In general, this variation in the patterns observed for our
392 two focal species points to the potential for character displacement to be context-dependent
393 (Eisen and Geber 2018; Roth-Monzon et al. 2020) and to occur via different phenotypic
394 pathways across communities (Germain et al. 2017). In particular, our results suggest that bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
395 changes in the volatile profile of C. cylindrica may result primarily from interactions with C.
396 unguiculata, which occurs in both types of sympatric communities. For C. cylindrica, indirect
397 interactions with the later-flowering Clarkia species in the four-species communities may not
398 affect the evolution of floral scent.
399 In contrast, C. unguiculata had greater emission rates of two aromatic compounds at both
400 types of sympatric communities, but lower emission rates of four monoterpenoids only at the
401 four-species sympatric communities. Similar patterns of intermediate or less displacement were
402 observed across different multispecies communities of freshwater fish (Roth-Monzon et al.
403 2020), which suggests that evolution in these communities likely occurs in response to multiple
404 species interactions. Because C. unguiculata is the earliest Clarkia species to flower in the region
405 (Moeller 2004; Singh 2014), its higher total scent emission (see Figure 2) may serve to attract
406 scarce pollinators at the beginning of the flowering season (Filella et al. 2013). However, the
407 observed decrease in monoterpenoid emissions in the four-species communities suggests that C.
408 unguiculata may invest less in pollinator attraction if, like other species of Clarkia (Moeller
409 2004), it experiences facilitation in these communities.
410 Multi-modal character displacement: synergy of changes in floral size & scent. Because
411 pollinators often exhibit responses to combinations of visual and olfactory traits (Leonard et al.
412 2011), we conducted the first test for character displacement in multi-modal floral signals
413 (Figure 4). Using estimates of volatile emission rates that were standardized by floral fresh mass,
414 we found that changes in the floral scent of C. unguiculata were not associated with changes in
415 flower size, while increases in the emission of floral scent of C. cylindrica were related to
416 increases in flower size. We hypothesize that this pattern results from differences in the floral
417 parts that produce these compounds (Effmert et al. 2006). Our floral dissections suggest that the bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
418 character displacement compounds in C. cylindrica are produced in both the petals and the
419 reproductive parts, which is consistent with a correlation between flower size and floral scent. In
420 contrast, the character displacement compounds in C. unguiculata were present more often in the
421 reproductive parts relative to the petals (Appendix 1), which is consistent with a change in floral
422 volatiles that was independent of a change in flower size. The differences in these patterns
423 highlight that the complexity of floral scent can extend beyond the quantitative and qualitative
424 composition of a scent bouquet to include spatial variation in the emission of volatiles across
425 tissue types (Friberg et al. 2013; Burdon et al. 2015; Martin et al. 2017).
426 These differences in the floral sources of the volatiles that change across the community
427 types may signify differences in their functions. For C. cylindrica, the increases in both size and
428 volatile emissions in both petals and reproductive parts may serve to increase overall pollinator
429 attraction in sympatry. Increases in flower size or scent emission have been linked to increased
430 pollinator attraction and plant reproductive success in multiple insect-pollinated systems (Conner
431 and Rush 1996; Miyake and Yafuso 2003; Majetic et al. 2009; Sandring and Ågren 2009;
432 Parachnowitsch et al. 2012), although most studies have not tested for concurrent changes in
433 both traits (but see Parachnowitsch et al. 2012). For C. unguiculata, scent emission in the
434 reproductive tissues may serve to cue pollinators to the precise location of the reproductive parts
435 (Dötterl and Jürgens 2005; Burdon et al. 2015). Because the solitary bees that specialize on
436 Clarkia forage for pollen (MacSwain et al. 1973), volatiles emitted in the reproductive tissues
437 also indicate the location of the primary rewards for this species. After becoming attracted to a
438 flower, bees can use pollen odors, which are often a distinct subset of the floral bouquet (Jürgens
439 and Dötterl 2004; Effmert et al. 2006), to orient more specifically to the source of pollen
440 (Dobson et al. 1996, 1999). Here, the decrease in floral volatiles that are putatively produced in bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
441 the reproductive organs in four-species communities suggests that C. unguiculata may invest less
442 not only in pollinator attraction as described above, but more specifically in provisioning
443 pollinators with pollen where the community of congeners may facilitate joint pollinator
444 attraction or maintenance (Moeller 2004). This hypothesis could be tested via additional analysis
445 of the pollen volatiles in C. unguiculata, and with pollinator behavior assays (see below).
446 Future directions. This study yielded a pattern of trait variation that is consistent with character
447 displacement, but additional work is needed to rule out alternative hypotheses (Schluter and
448 McPhail 1992). In particular, it is critical to determine if this variation in scent has functional
449 consequences for pollinator behavior. Given that the volatiles that mediate pollinator behavior
450 are often a subset of all volatiles emitted by a plant (reviewed in Junker and Blüthgen 2010),
451 pollinators may not respond to the specific changes observed in floral scent profiles across
452 community types. Experimental assays of pollinator behavior can be used to determine if these
453 shifts in volatiles affect pollinator attraction or constancy, or if they are non-functional. The
454 potential effects of variation in C. unguiculata volatiles on honey bees and bumblebees could be
455 tested in a controlled environment (e.g., Burger et al. 2012; Peter and Johnson 2014). However, a
456 comprehensive assessment of the functionality of floral scent variation in Clarkia would need to
457 be field-based, as lab experiments with the solitary bees that pollinate both species are not
458 tractable.
459 More broadly, the results of this study highlight the need to continue to integrate
460 chemical phenotypes into the study of floral trait evolution (Leonard et al. 2011; Junker and
461 Parachnowitsch 2015). In combination with visual traits, floral scent can affect species
462 interactions at multiple scales, from specifying highly specialized interactions (e.g., Peakall and
463 Whitehead 2014; Whitehead et al. 2015) to contributing to the structure of complex plant- bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
464 pollinator interaction networks (Kantsa et al. 2018, 2019). In this study, by testing for character
465 displacement in multiple trait modalities across communities that contain different numbers of
466 interacting species, we have generated new insights into the context-dependency of character
467 displacement, which may occur through multiple pathways. Moving forward, systems that
468 exhibit variation in both floral scent and species interactions across communities (e.g., Friberg et
469 al. 2019) provide opportunities to study the interplay between complex trait evolution and
470 species interactions, which can generate insight into the repeatability of evolutionary change
471 across variable ecological communities.
472 Acknowledgements: We thank Geoff Broadhead and Amy Wruck for assistance in the lab and
473 greenhouse. We thank Erika Mudrak from the Cornell Statistical Consulting Center and Diane
474 Campbell for advice on statistical analyses. Anurag Agrawal provided insightful comments on
475 drafts of the manuscript. This work was funded by a Graduate Research Fellowship (DGE-
476 1650441) from the US National Science Foundation, and a Cornell Atkinson Center for
477 Sustainability Sustainable Biodiversity Fund grant to KEE, by NSF DEB-1754299 to MAG, and
478 by NSF DEB-1342792 to RAR.
479 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
480 Table 1. The three compounds with the strongest positive and negative correlations with the first 481 two CAP axes. For CAP1, nine additional compounds (two “green leafy volatiles,” five 482 monoterpenoids, one sesquiterpenoid, and one aromatic compound) were significantly positively 483 correlated with this axis, and three additional sesquiterpenoids and two aromatic compounds 484 exhibited significant negative correlations (r > 0.15; Table S3). For CAP2, two additional 485 monoterpenoids and one aromatic compound were also positively correlated with this axis, and 486 three additional monoterpenes exhibited significant negative correlations (r > 0.15; Table S3). A 487 false discovery rate correction for conducting 54 tests was applied to all P values (see Methods). 488 Axis and compound r P CAP1 (Z)-β-ocimene 0.528 1.51 x 10-19 (E)-β-ocimene 0.526 1.70 x 10-19 2-amino phenyl ethanone 0.481 4.60 x 10-16 intermedeol -0.646 1.73 x 10-31 β-cadinene -0.641 2.87 x 10-31 -18 unknown C15H24 #3 -0.516 1.04 x 10 CAP2 (E)-β-ocimene 0.457 1.26 x 10-13 (Z)-β-ocimene 0.444 5.16 x 10-13 (E)-2-6-dimethyl-1,3,5,7- octatetraene 0.329 4.12 x 10-7 (Z)-3-hexenyl acetate -0.349 6.44 x 10-8 (Z)-3-hexen-1-ol -0.275 3.52 x 10-5 sabinene hydrate -0.275 3.52 x 10-5 489 490 491 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license. 4
2-amino-phenyl ethanone (E)-β-ocimene
2 2-6-dimethyl-1-3-5-7-octatetraene (E) (Z)-β-ocimene
methyl nicotinate borneol
β-longipinene myroxide γ-cadinene trans cinnamic aldehyde (Z)-3-hexen-1-ol unknown C H #2 β-cadinene 15 24 0 CAP2 β-myrcene intermedeol veratrole unknown C H #3 15 24 α-bergamotene (Z)-3-hexenyl acetate
γ-terpinene sabinene hydrate β-pinene
-2 α-pinene
C. cylindrica C. unguiculata Four Four Two Two One One
-6 -4 -2 0 2 4 6 CAP1 492
493 Figure 1. Canonical analysis of principle coordinates separating volatile emissions from the two 494 species and three community types. The different species and community type combinations are 495 indicated by color and symbol type. The loadings are shown with arrows for compounds with the 496 largest positive and negative correlations (r) with CAP axis 1 (underlined compounds), and for 497 compounds with the largest positive and negative correlations (r) with CAP axis 2 (italicized 498 compounds). CAP axis 1 strongly separates the species, as C. cylindrica have lower values and 499 C. unguiculata have higher values. CAP axis 2 primarily differentiates the community types of 500 C. unguiculata, as individuals from one-species communities had generally low values 501 (centroid= -1.212), individuals from two-species communities had intermediate values (centroid: 502 -0.061), and individuals from four-species communities had high values (centroid: 1.330). 503 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
A: Monoterpenoids B: Sesquiterpenoids A A A A B AB b 0.125 b b b
0.6 0.100 b
0.075 0.4
b Emission rate
Emission rate 0.050 a 0.2 a a a 0.025 a a
One Two Four One Two Four C: “Green leafy volatiles” D: Aromatics B B A A B A 0.25 b b
0.015 0.20 b a
0.012 0.15
0.009 0.10 Emission rate Emission rate a a a a a a 0.05 0.006 a a
0.00 0.003 One Two Four One Two Four CommunityCommunity Typetype CommunityCommunity Typetype
C. cylindrica C. unguiculata 504 505 Figure 2. Emission rates (raw values; µg scent/g fresh floral mass/hour) of monoterpenoids (A), 506 sesquiterpenoids (B), “green leafy volatiles” (C), and aromatic compounds (D) by C. cylindrica 507 (navy blue) and C. unguiculata (dark pink). Lowercase letters above each point indicate whether 508 emission rates differed between the species within that community type; the letter b indicates the 509 species with the higher emission rate. Uppercase letters above each set of points at a community 510 type indicate whether that difference is the same or different from the differences at other 511 community types; differences with the letter A in italics are not significantly different from zero, 512 and differences with the letters B or C are larger than differences with the letter A. Emission 513 rates of monoterpenoids were higher in C. unguiculata (A). Clarkia cylindrica produced more 514 sesquiterpenes than C. unguiculata at all community types (B), but the difference between the 515 species was significantly larger at one-species communities than at two-species communities. 516 Clarkia unguiculata produced more “green leafy volatiles” than C. cylindrica at one-species bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
517 communities and at two-species communities, but emission rates at four-species communities 518 were equivalent (C). Clarkia cylindrica had substantially higher emission rates of aromatic 519 compounds at two-species communities (D). Note the differences in scale for the y-axes across 520 all panels. bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
A: 2 amino O B: methyl O C: α-pinene phenyl ethanone H nicotinate N H N O A AB B A B AB 0.0015 B B A b 0.004 b b b b
0.0010 0.003 0.010 b
0.002
Emission rate 0.0005 a 0.005 Emission rate 0.001 a b a a a a a a a a a
0.0000 0.000 0.000
One Two Four One Two Four One Two Four
D: β-pinene CH2 E: sabinene OH F: γ-terpinene CH hydrate H
B B A 8e-04 B B A 0.003 B B A b b b b 0.00100 b 6e-04 b 0.002 0.00075
4e-04
0.00050
Emission rate 0.001
2e-04 Emission rate 0.00025 b a a a a a a a a a a a
0.00000 0e+00 0.000 One Two Four One Two Four One Two Four
O G: (E)-cinnamic H: veratrole OCH3 I: (Z)-3-hexenyl O H H aldehyde acetate O H OCH3
A B AB 0.003 A C B B B A a b b b 0.20
1e-03 b a 0.002 0.15 b b
0.10 5e-04 Emission rate 0.001 a
Emission rate 0.05 a a a a a a a a a
0e+00 0.000 0.00
One Two Four One Two Four One Two Four CommunityCommunity type type CommunityCommunity typetype CommunityCommunity type
C. cylindrica C. unguiculata 521 Figure 3. Emission rates (raw values; µg scent/g fresh floral mass/hour) of the nine compounds 522 that were significantly correlated with one or both of the first two CAP axes and also had a 523 significant species x community type interaction. Lowercase letters above each point indicate 524 whether emission rates differed between the species within that community type; the letter b 525 indicates the species with the higher emission rate. Uppercase letters above each set of points at a bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
526 community type indicate whether that difference is the same or different from the differences at 527 other community types; differences with the letter A in italics are not significantly different from 528 zero, and differences with the letters B or C are larger than differences with the letter A. 2- 529 amino-phenyl ethenone (A) and methyl nicotinate (B) were primarily emitted by C. unguiculata 530 (dark pink) and increased in two- and four-species communities. α-pinene (C), β-pinene (D), 531 sabinene hydrate (E), and γ-terpinene (F) were also primarily emitted by C. unguiculata and 532 decreased in four-species communities. (E)-cinnamic aldehyde (G) and veratrole (H) were 533 exclusively emitted by C. cylindrica (navy blue) and the emission rates of these compounds 534 increased in two- and four-species communities. (Z)-3-hexenyl acetate (I) was emitted at higher 535 rates by C. unguiculata at one- and two-species communities, and by C. cylindrica at four- 536 species communities. Note the differences in scale for the y-axes across all panels. 537 538 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
One-species communities Two-species communities Four-species communities
C. unguiculata
C. cylindrica
539 540 Figure 4. Schematic showing the relative changes in flower size (based on measurements of petal 541 area in Eisen & Geber 2018) and the species-specific floral scent compounds that showed 542 patterns consistent with character displacement (C. unguiculata: 2-amino phenyl ethenone, α- 543 pinene, β-pinene, sabinene hydrate, γ-terpinene, and methyl nicotinate; C. cylindrica: (E)- 544 cinnamic aldehyde and veratrole). Drawings of flowers and chemical compounds (molecules that 545 are representative of the suites of compounds that responded in each species) are scaled 546 proportionally both between the species and across the community types. Flower size of C. 547 unguiculata is similar across community types and is slightly smaller than the flower size of C. 548 cylindrica at one-species communities. Floral scent emission rates of C. unguiculata are similar 549 at one- and two-species communities, and emission rates at four-species communities are about 550 0.45 times emission rates at one species-communities (a decrease in scent emission in four- 551 species communities). Emission rates of floral scent in C. unguiculata are one or two orders of 552 magnitude higher than emission rates in C. cylindrica. Flower size of C. cylindrica at two- 553 species communities is 1.7 times larger than flower size at one-species communities, and flower 554 size at four-species communities is 1.25 times larger than at one-species communities. Floral 555 scent emission rates of C. cylindrica are 7.2 times larger at two-species communities relative to 556 one-species communities, and 4.6 times larger at four-species communities relative to one- 557 species communities. 558 bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
559 References
560 Andrews, E. S., N. Theis, and L. S. Adler. 2007. Pollinator and herbivore attraction to Cucurbita
561 floral volatiles. Journal of Chemical Ecology 33:1682–1691.
562 Bates, D., M. Mächler, B. M. Bolker, and S. C. Walker. 2015. Fitting linear mixed-effects
563 models using lme4. Journal of Statistical Software 67:1–48.
564 Beans, C. M. 2014. The case for character displacement in plants. Ecology and Evolution 4:852–
565 65.
566 Becklin, K. M., G. Gamez, B. Uelk, R. A. Raguso, and C. Galen. 2011. Soil fungal effects on
567 floral signals, rewards, and aboveground interactions in an alpine pollination web. American
568 Journal of Botany 98:1299–1308.
569 Benkman, C. W. 2013. Biotic interaction strength and the intensity of selection. Ecology Letters
570 16:1054–1060.
571 Bischoff, M., A. Jürgens, and D. R. Campbell. 2014. Floral scent in natural hybrids of Ipomopsis
572 (Polemoniaceae) and their parental species. Annals of Botany 113:533–544.
573 Brown, W. L., and E. O. Wilson. 1956. Character displacement. Systematic Zoology 5:49–64.
574 Burdon, R. C. F., R. A. Raguso, A. Kessler, and A. L. Parachnowitsch. 2015. Spatiotemporal
575 floral scent variation of Penstemon digitalis. Journal of Chemical Ecology 41:641–650.
576 Burger, H., S. Dötterl, C. M. Häberlein, S. Schulz, and M. Ayasse. 2012. An arthropod deterrent
577 attracts specialised bees to their host plants. Oecologia 168:727–736.
578 Chapurlat, E., J. Agren, J. Anderson, M. Friberg, and N. Sletvold. 2019. Conflicting selection on
579 floral sent emission in the orchid Gymnadenia conopsea. New Phytologist 222:2009–2022.
580 Chapurlat, E., J. Anderson, J. Ågren, M. Friberg, and N. Sletvold. 2018. Diel pattern of floral
581 scent emission matches the relative importance of diurnal and nocturnal pollinators in bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
582 populations of Gymnadenia conopsea. Annals of Botany 121:711–721.
583 Conner, J. K., and S. Rush. 1996. Effects of flower size and number on pollinator visitation to
584 wild radish, Raphanus raphanistrum. Oecologia 105:509–516.
585 Darwin, C. 1859. The origin of species. John Murray, London.
586 Dobson, H. E. M. 2006. Relationship between floral fragrance composition and type of
587 pollinator. Pages 147–198 in E. Pichersky and N. Dudareva, eds. Biology of Floral Scent.
588 Chapman & Hall.
589 Dobson, H. E. M., E. M. Danielson, and I. D. Van Wesep. 1999. Pollen odor chemicals as
590 modulators of bumble bee foraging on Rosa rugosa Thunb. (Rosaceae). Plant Species Biology
591 14:153–166.
592 Dobson, H. E. M., I. Groth, and G. Bergström. 1996. Pollen advertisement: chemical contrasts
593 between whole-flower and pollen odors. American Journal of Botany 83:877–885.
594 Dötterl, S., and A. Jürgens. 2005. Spatial fragrance patterns in flowers of Silene latifolia: Lilac
595 compounds as olfactory nectar guides? Plant Systematics and Evolution 255:99–109.
596 Dudareva, N., and E. Pichersky. 2006. Floral scent metabolic pathways: Their regulation and
597 evolution. Pages 55–78 in E. Pichersky and N. Dudareva, eds. Biology of Floral Scent. Chapman
598 & Hall.
599 Effmert, U., D. Buss, D. Rohrbeck, and B. Piechulla. 2006. Localization of the synthesis and
600 emission of scent compounds within the flower. Pages 105–124 in N. Dudareva and E.
601 Pichersky, eds. Biology of Floral Scent. CRC Press, Boca Raton, FL.
602 Eisen, K. E., and M. A. Geber. 2018. Ecological sorting and character displacement contribute to
603 the structure of communities of Clarkia species. Journal of Evolutionary Biology 31:1440–1458.
604 Filella, I., C. Primante, J. Llusià, A. M. Martín González, R. Seco, G. Farré-Armengol, A. bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
605 Rodrigo, et al. 2013. Floral advertisement scent in a changing plant-pollinators market. Scientific
606 Reports 3:1–6.
607 Friberg, M., C. Schwind, P. R. Guimarães, R. A. Raguso, and J. N. Thompson. 2019. Extreme
608 diversification of floral volatiles within and among species of Lithophragma (Saxifragaceae).
609 Proceedings of the National Academy of Sciences 201809007.
610 Friberg, M., C. Schwind, R. A. Raguso, and J. N. Thompson. 2013. Extreme divergence in floral
611 scent among woodland star species (Lithophragma spp.) pollinated by floral parasites. Annals of
612 Botany 111:539–550.
613 Germain, R. M., J. L. Williams, D. Schluter, and A. L. Angert. 2017. Moving character
614 displacement beyond characters using contemporary coexistence theory. Trends in Ecology &
615 Evolution 33:74–84.
616 Gould, B., D. A. Moeller, V. M. Eckhart, P. Tiffin, E. Fabio, and M. A. Geber. 2014. Local
617 adaptation and range boundary formation in response to complex environmental gradients across
618 the geographical range of Clarkia xantiana ssp . xantiana. Journal of Ecology 102:95–107.
619 Grant, P. R. 2017. Ecology and Evolution of Darwin’s Finches. Princeton University Press,
620 Princeton, New Jersey, USA.
621 Grant, P. R., and B. R. Grant. 2008. How and why species multiply: the radiation of Darwin’s
622 finches. Princeton University Press, Princeton, NJ.
623 Gross, K., M. Sun, and F. P. Schiestl. 2016. Why do floral perfumes become different? Region-
624 specific selection on floral scent in a terrestrial orchid. PLoS ONE 11:e0147975.
625 Hopkins, R., and M. D. Rausher. 2012. Pollinator-mediated selection on flower color allele
626 drives reinforcement. Science 335:1090–1092.
627 Jonas, C. S., and M. A. Geber. 1999. Variation among populations of Clarkia unguiculata bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
628 (Onagraceae) along altitudinal and latitudinal gradients. American Journal of Botany 86:333–
629 343.
630 Junker, R. R., and N. Blüthgen. 2010. Floral scents repel facultative flower visitors, but attract
631 obligate ones. Annals of Botany 105:777–782.
632 Junker, R. R., N. Höcherl, and N. Blüthgen. 2010. Responses to olfactory signals reflect network
633 structure of flower-visitor interactions. Journal of Animal Ecology 79:818–823.
634 Junker, R. R., and A. L. Parachnowitsch. 2015. Working towards a holistic view on flower traits-
635 how floral scents mediate plant-animal interactions in concert with other floral characters.
636 Journal of the Indian Institute of Science 95:43–67.
637 Jürgens, A., and S. Dötterl. 2004. Chemical composition of anther volatiles in Ranunculaceae:
638 Genera-specific profiles in Anemone, Aquilegia, Caltha, Pulsatilla, Ranunculus, and Trollius
639 species. American Journal of Botany 91:1969–1980.
640 Kantsa, A., R. A. Raguso, A. G. Dyer, J. M. Olesen, T. Tscheulin, and T. Petanidou. 2018.
641 Disentangling the role of floral sensory stimuli in pollination networks. Nature Communications
642 9:1–25.
643 Kantsa, A., R. A. Raguso, T. Lekkas, O. I. Kalantzi, and T. Petanidou. 2019. Floral volatiles and
644 visitors: A meta-network of associations in a natural community. Journal of Ecology 107:2574–
645 2586.
646 Knudsen, J. T., R. Eriksson, J. Gershenzon, and B. Ståhl. 2006. Diversity and distribution of
647 floral scent. The Botanical Review 72:1–120.
648 Kuznetsova, A., P. Brockhoff, and R. Christensen. 2015. Package “lmerTest.” R package
649 version.
650 Lack, D. 1945. The Galapagos finches (Geospizinae): a study in variation. California Academy bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
651 of Sciences.
652 Larue, A. A. C., R. A. Raguso, and R. R. Junker. 2016. Experimental manipulation of floral scent
653 bouquets restructures flower-visitor interactions in the field. Journal of Animal Ecology 85:396–
654 408.
655 Lemmon, E. M., and A. R. Lemmon. 2010. Reinforcement in chorus frogs: Lifetime fitness
656 estimates including intrinsic natural selection and sexual selection against hybrids. Evolution
657 64:1748–1761.
658 Lenth, R. 2019. emmeans: Estimated Marginal Means, aka Least-Squares Means. R package
659 version 1.3.3.
660 Leonard, A. S., A. Dornhaus, and D. R. Papaj. 2011. Forget-me-not: Complex floral displays,
661 inter-signal interactions, and pollinator cognition. Current Zoology 57:215–224.
662 Levin, D. A. 1985. Reproductive character displacement in Phlox. Evolution 39:1275–1281.
663 Lewis, H. 1953. The mechanism of evolution in the genus Clarkia. Evolution 7:1–20.
664 ———. 1961. Experimental sympatric populations of Clarkia. The American Naturalist 95:155–
665 168.
666 Lewis, H., and M. E. Lewis. 1955. The genus Clarkia. University of California Publications in
667 Botany 20:241–392.
668 Losos, J. B. 2011. Convergence, adaptation, and constraint. Evolution 65:1827–1840.
669 MacSwain, J., P. H. Raven, and R. Thorp. 1973. Comparative behavior of bees and Onagraceae.
670 IV. Clarkia bees of the western United States. University of California Publications in
671 Entomology 70:1–80.
672 Majetic, C. J., R. A. Raguso, and T. L. Ashman. 2009. The sweet smell of success: Floral scent
673 affects pollinator attraction and seed fitness in Hesperis matronalis. Functional Ecology 23:480– bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
674 487.
675 Martin, K. R., M. Moré, J. Hipólito, S. Charlemagne, B. O. Schlumpberger, and R. A. Raguso.
676 2017. Spatial and temporal variation in volatile composition suggests olfactory division of labor
677 within the trap flowers of Aristolochia gigantea. Flora: Morphology, Distribution, Functional
678 Ecology of Plants 232:153–168.
679 Mayfield, M. M., and D. B. Stouffer. 2017. Higher-order interactions capture unexplained
680 complexity in diverse communities. Nature Ecology & Evolution 1:0062.
681 Miller, T. E., E. R. Moran, and C. P. TerHorst. 2014a. Rethinking niche evolution: Experiments
682 with natural communities of Protozoa in pitcher plants. The American Naturalist 184:277–283.
683 Miller, T. J., R. A. Raguso, and K. M. Kay. 2014b. Novel adaptation to hawkmoth pollinators in
684 Clarkia reduces efficiency, not attraction of diurnal visitors. Annals of botany 113:317–29.
685 Miyake, T., and M. Yafuso. 2003. Floral scents affect reproductive success in fly-pollinated
686 Alocasia odora (Araceae). American Journal of Botany 90:370–376.
687 Moeller, D. A. 2004. Facilitative interactions among plants via shared pollinators. Ecology
688 85:3289–3301.
689 Muchhala, N., S. Johnsen, and S. Smith. 2014. Competition for hummingbird pollination shapes
690 flower color variation in Andean Solanaceae. Evolution 68:2275–2286.
691 Norton, N. A., M. T. R. Fernando, C. R. Herlihy, and J. W. Busch. 2015. Reproductive character
692 displacement shapes a spatially structured petal color polymorphism in Leavenworthia stylosa.
693 Evolution 69:1191–1207.
694 Oksanen, J., F. G. Blanchet, M. Friendly, R. Kindt, P. Legendre, D. McGlinn, P. R. Minchin, et
695 al. 2018. vegan: Community Ecology Package. R package version 2.5-2.
696 Ollerton, J., R. Winfree, and S. Tarrant. 2011. How many flowering plants are pollinated by bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
697 animals? Oikos 120:321–326.
698 Parachnowitsch, A. L., R. A. Raguso, and A. Kessler. 2012. Phenotypic selection to increase
699 floral scent emission, but not flower size or colour in bee-pollinated Penstemon digitalis. New
700 Phytologist 195:667–675.
701 Peakall, R., and M. R. Whitehead. 2014. Floral odour chemistry defines species boundaries and
702 underpins strong reproductive isolation in sexually deceptive orchids. Annals of Botany
703 113:341–355.
704 Peter, C. I., and S. D. Johnson. 2014. A pollinator shift explains floral divergence in an orchid
705 species complex in South Africa. Annals of Botany 113:277–288.
706 R Core Team. 2018. R: A language and environment for statistical computing. R Foundation for
707 Statistical Computing, Vienna, Austria.
708 Raguso, R. A. 2008a. Start making scents: The challenge of integrating chemistry into
709 pollination ecology. Entomologia Experimentalis et Applicata 128:196–207.
710 ———. 2008b. Wake up and smell the roses: The ecology and evolution of floral scent. Annual
711 Review of Ecology, Evolution, and Systematics 39:549–569.
712 ———. 2014. A wrapped bouquet: the untapped potential of floral chemistry. Journal of
713 Chemical Ecology 40:412–3.
714 Roth-Monzon, A. J., M. C. Belk, J. J. Zuniga-Vega, and J. B. Johnson. 2020. Beyond pairwise
715 interactions: multispecies character displacement in Mexican freshwater fish communities.
716 American Naturalist doi: 10.1086/708513.
717 Sandring, S., and J. Ågren. 2009. Pollinator-mediated selection on floral display and flowering
718 time in the perennial herb Arabidopsis lyrata. Evolution 63:1292–1300.
719 Sargent, R. D., and D. D. Ackerly. 2008. Plant-pollinator interactions and the assembly of plant bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
720 communities. Trends in Ecology & Evolution 23:123–30.
721 Scala, A., S. Allmann, R. Mirabella, M. A. Haring, and R. C. Schuurink. 2013. Green leaf
722 volatiles: A plant’s multifunctional weapon against herbivores and pathogens. International
723 Journal of Molecular Sciences 14:17781–17811.
724 Schiestl, F. P. 2010. The evolution of floral scent and insect chemical communication. Ecology
725 Letters 13:643–656.
726 Schiestl, F. P. 2015. Ecology and evolution of floral volatile- mediated information transfer in
727 plants. New Phytologist 206:571–577.
728 Schlumpberger, B. O., and R. A. Raguso. 2008. Geographic variation in floral scent of
729 Echinopsis ancistrophora (Cactaceae); evidence for constraints on hawkmoth attraction. Oikos
730 117:801–814.
731 Schluter, D. 2000. The ecology of adaptive radiation. Oxford University Press.
732 Schluter, D., and J. D. McPhail. 1992. Ecological character displacement and speciation in
733 sticklebacks. The American Naturalist 140:85–108.
734 Singh, I. 2014. Pollination interaction networks between Clarkia (Onagraceae) species and their
735 pollinators in the Southern Sierra Nevada, California. Cornell University.
736 Soler, C., M. Proffit, C. Chen, and M. Hossaert-McKey. 2010. Private channels in plant-
737 pollinator mutualisms. Plant Signaling and Behavior 5:893–895.
738 Stuart, Y. E., and J. B. Losos. 2013. Ecological character displacement: Glass half full or half
739 empty? Trends in Ecology and Evolution 28:402–408.
740 Suinyuy, T. N., J. S. Donaldson, and S. D. Johnson. 2012. Geographical variation in cone
741 volatile composition among populations of the African cycad Encephalartos villosus. Biological
742 Journal of the Linnean Society 106:514–527. bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
743 Svensson, G. P., M. O. Hickman, S. Bartram, W. Boland, O. Pellmyr, and R. A. Raguso. 2005.
744 Chemistry and geographic variation of floral scent in Yucca filamentosa (Agavaceae). American
745 Journal of Botany 92:1624–1631.
746 TerHorst, C. P., J. A. Lau, I. A. Cooper, K. R. Keller, R. J. La Rosa, A. M. Royer, E. H.
747 Schultheis, et al. 2015. Quantifying non-additive selection caused by indirect ecological effects.
748 Ecology 96:2360–2369.
749 TerHorst, C. P., P. C. Zee, K. D. Heath, T. E. Miller, A. I. Pastore, S. Patel, S. J. Schreiber, et al.
750 2018. Evolution in a community context: Trait responses to multiple species interactions. The
751 American Naturalist 191:368–380.
752 Theis, N. 2006. Fragrance of Canada thistle (Cirsium arvense) attracts both floral herbivores and
753 pollinators. Journal of Chemical Ecology 32:917–927.
754 Valdivia, C. E., and H. M. Niemeyer. 2006. Do pollinators simultaneously select for
755 inflorescence size and amount of floral scents? An experimental assessment on Escallonia
756 myrtoidea. Austral Ecology 31:897–903.
757 Visser, J., and D. Ave. 1978. General green leaf volatiles in the olfactory orientation of the
758 Colorado beetle, Lerptinotarsa decemlineata. Entomologia Experimentalis et Applicata 24:738–
759 749.
760 Waelti, M. O., J. K. Muhlemann, A. Widmer, and F. P. Schiestl. 2008. Floral odour and
761 reproductive isolation in two species of Silene. Journal of Evolutionary Biology 21:111–121.
762 Walsh, M. R. 2013. The evolutionary consequences of indirect effects. Trends in Ecology and
763 Evolution 28:23–29.
764 Waser, N. M. 1986. Flower constancy: definition, cause, and measurement. The American
765 Naturalist 127:593–603. bioRxiv preprint doi: https://doi.org/10.1101/2020.04.02.022004; this version posted April 3, 2020. The copyright holder for this preprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made available under aCC-BY-NC-ND 4.0 International license.
766 Waser, N. M., L. Chittka, M. V Price, N. M. Williams, and J. Ollerton. 1996. Generalization in
767 pollination systems, and why it matters. Ecology 77:1043–1060.
768 Whitehead, M. R., C. C. Linde, and R. Peakall. 2015. Pollination by sexual deception promotes
769 outcrossing and mate diversity in self-compatible clonal orchids. Journal of Evolutionary
770 Biology 28:1526–1541.
771